diff --git a/jenkins/L0_Test.groovy b/jenkins/L0_Test.groovy
index c96dc010583..949209fa205 100644
--- a/jenkins/L0_Test.groovy
+++ b/jenkins/L0_Test.groovy
@@ -261,7 +261,7 @@ def runLLMTestlistOnSlurm(pipeline, platform, testList, config=VANILLA_CONFIG, p
             }
 
             if (CloudManager.isNodeOnline(nodeName)) {
-                def dockerArgs = "--gpus ${gpuCount} --cap-add=SYS_ADMIN --ipc=host --security-opt seccomp=unconfined  -u root:root -v /home/scratch.trt_llm_data:/scratch.trt_llm_data:ro -v /tmp/ccache:${CCACHE_DIR}:rw -v /tmp/pipcache/http-v2:/root/.cache/pip/http-v2:rw --cap-add syslog"
+                def dockerArgs = "--gpus ${gpuCount} --cap-add=SYS_ADMIN --ipc=host --security-opt seccomp=unconfined  -u root:root -v /home/scratch.trt_llm_data:/scratch.trt_llm_data:ro -v /tmp/ccache:${CCACHE_DIR}:rw -v /tmp/pipcache/http-v2:/root/.cache/pip/http-v2:rw --cap-add syslog -e NVIDIA_IMEX_CHANNELS=0"
                 slurmRunner = runInDockerOnNodeMultiStage(LLM_DOCKER_IMAGE, nodeName, dockerArgs, false)
                 executeLLMTestOnSlurm(pipeline, platform, testList, config, perfMode, stageName, splitId, splits, skipInstallWheel, cpver, slurmRunner)
             } else {
@@ -362,6 +362,7 @@ def runLLMTestlistOnSlurm_MultiNodes(pipeline, platform, testList, config=VANILL
                     "--container-image=${container}",
                     "--container-workdir=/home/svc_tensorrt/bloom/scripts",
                     "--container-mounts=${mounts}",
+                    "--container-env=NVIDIA_IMEX_CHANNELS"
                 ].join(" ")
 
                 def scriptLaunch = "/home/svc_tensorrt/bloom/scripts/${jobUID}/slurm_launch.sh"
@@ -382,6 +383,7 @@ def runLLMTestlistOnSlurm_MultiNodes(pipeline, platform, testList, config=VANILL
                     export perfMode=$perfMode
                     export resourcePathNode=$resourcePathNode
                     export MODEL_CACHE_DIR=$MODEL_CACHE_DIR
+                    export NVIDIA_IMEX_CHANNELS=0
                     chmod +x ${scriptRunNode}
                     ${srunCmd}
                 """.stripIndent()
diff --git a/tests/integration/defs/accuracy/test_llm_api_pytorch.py b/tests/integration/defs/accuracy/test_llm_api_pytorch.py
index 45c67a63112..acc48855da0 100644
--- a/tests/integration/defs/accuracy/test_llm_api_pytorch.py
+++ b/tests/integration/defs/accuracy/test_llm_api_pytorch.py
@@ -25,8 +25,7 @@
                                  TorchCompileConfig)
 from tensorrt_llm.quantization import QuantAlgo
 
-from ..conftest import (llm_models_root, parametrize_with_ids,
-                        skip_device_contain_gb200, skip_no_hopper,
+from ..conftest import (llm_models_root, parametrize_with_ids, skip_no_hopper,
                         skip_post_blackwell, skip_pre_ada, skip_pre_blackwell,
                         skip_pre_hopper)
 from .accuracy_core import (GSM8K, MMLU, CnnDailymail, GPQADiamond,
@@ -85,9 +84,7 @@ def test_chunked_prefill(self, attn_backend):
             task.evaluate(llm)
 
     @pytest.mark.skip_less_device_memory(32000)
-    @parametrize_with_ids(
-        "torch_compile",
-        [False, pytest.param(True, marks=skip_device_contain_gb200)])
+    @parametrize_with_ids("torch_compile", [False, True])
     @parametrize_with_ids("attn_backend", ["TRTLLM", "FLASHINFER"])
     def test_bfloat16(self, attn_backend, torch_compile):
         torch_compile_config = TorchCompileConfig(
@@ -103,9 +100,7 @@ def test_bfloat16(self, attn_backend, torch_compile):
             task = GSM8K(self.MODEL_NAME)
             task.evaluate(llm)
 
-    @parametrize_with_ids(
-        "torch_compile",
-        [False, pytest.param(True, marks=skip_device_contain_gb200)])
+    @parametrize_with_ids("torch_compile", [False, True])
     @parametrize_with_ids("attn_backend", ["TRTLLM", "FLASHINFER"])
     @pytest.mark.parametrize("tp_size,pp_size", [(4, 1), (2, 2), (1, 4)],
                              ids=["tp4", "tp2pp2", "pp4"])
@@ -133,9 +128,7 @@ def test_bfloat16_4gpus(self, tp_size, pp_size, attn_backend,
             task.evaluate(llm)
 
     @skip_pre_ada
-    @parametrize_with_ids(
-        "torch_compile",
-        [False, pytest.param(True, marks=skip_device_contain_gb200)])
+    @parametrize_with_ids("torch_compile", [False, True])
     @parametrize_with_ids("attn_backend", ["TRTLLM", "FLASHINFER"])
     @parametrize_with_ids("fp8kv", [False, True])
     def test_fp8(self, fp8kv, attn_backend, torch_compile):
@@ -158,9 +151,7 @@ def test_fp8(self, fp8kv, attn_backend, torch_compile):
             task.evaluate(llm)
 
     @skip_pre_ada
-    @parametrize_with_ids(
-        "torch_compile",
-        [False, pytest.param(True, marks=skip_device_contain_gb200)])
+    @parametrize_with_ids("torch_compile", [False, True])
     @parametrize_with_ids("attn_backend", ["TRTLLM", "FLASHINFER"])
     @parametrize_with_ids("fp8kv", [False, True])
     @pytest.mark.parametrize("tp_size,pp_size", [(4, 1), (2, 2), (1, 4)],
@@ -643,9 +634,7 @@ class TestDeepSeekV3Lite(LlmapiAccuracyTestHarness):
     MODEL_PATH = f"{llm_models_root()}/DeepSeek-V3-Lite/bf16"
 
     @pytest.mark.skip_less_device_memory(60000)
-    @parametrize_with_ids(
-        "torch_compile",
-        [False, pytest.param(True, marks=skip_device_contain_gb200)])
+    @parametrize_with_ids("torch_compile", [False, True])
     @parametrize_with_ids("attention_dp,cuda_graph,overlap_scheduler",
                           [(False, False, False), (True, False, False),
                            (False, True, False), (False, False, True),
@@ -679,9 +668,7 @@ def test_bfloat16(self, mtp_nextn, attention_dp, cuda_graph,
             task.evaluate(llm)
 
     @pytest.mark.skip_less_device(4)
-    @parametrize_with_ids(
-        "torch_compile",
-        [False, pytest.param(True, marks=skip_device_contain_gb200)])
+    @parametrize_with_ids("torch_compile", [False, True])
     @parametrize_with_ids("attention_dp,cuda_graph,overlap_scheduler",
                           [(False, False, False), (True, False, False),
                            (False, True, False), (False, False, True),
@@ -724,9 +711,7 @@ def test_bfloat16_4gpus(self, tp_size, pp_size, ep_size, mtp_nextn,
             task.evaluate(llm)
 
     @skip_no_hopper
-    @parametrize_with_ids(
-        "torch_compile",
-        [False, pytest.param(True, marks=skip_device_contain_gb200)])
+    @parametrize_with_ids("torch_compile", [False, True])
     @parametrize_with_ids("fp8kv,attention_dp,cuda_graph,overlap_scheduler",
                           [(False, False, False, False),
                            (True, False, False, False),
@@ -871,9 +856,7 @@ def test_fp8_block_scales_cuda_graph_padding_4gpus(self, mtp_nextn,
 
     @pytest.mark.skip_less_device(4)
     @skip_no_hopper
-    @parametrize_with_ids(
-        "torch_compile",
-        [False, pytest.param(True, marks=skip_device_contain_gb200)])
+    @parametrize_with_ids("torch_compile", [False, True])
     @parametrize_with_ids("fp8kv,attention_dp,cuda_graph,overlap_scheduler",
                           [(False, False, False, False),
                            (True, False, False, False),
@@ -1069,9 +1052,7 @@ def test_nvfp4_4gpus_online_eplb(self, fp8kv):
             task.evaluate(llm)
 
     @skip_pre_blackwell
-    @parametrize_with_ids(
-        "torch_compile",
-        [False, pytest.param(True, marks=skip_device_contain_gb200)])
+    @parametrize_with_ids("torch_compile", [False, True])
     @parametrize_with_ids("fp8kv,attention_dp,cuda_graph,overlap_scheduler",
                           [(False, False, False, False),
                            (True, False, False, False),
@@ -1113,9 +1094,7 @@ def test_nvfp4(self, fp8kv, attention_dp, cuda_graph, overlap_scheduler,
 
     @pytest.mark.skip_less_device(4)
     @skip_pre_blackwell
-    @parametrize_with_ids(
-        "torch_compile",
-        [False, pytest.param(True, marks=skip_device_contain_gb200)])
+    @parametrize_with_ids("torch_compile", [False, True])
     @parametrize_with_ids("fp8kv,attention_dp,cuda_graph,overlap_scheduler",
                           [(False, False, False, False),
                            (True, False, False, False),
@@ -1351,8 +1330,7 @@ class TestDeepSeekR1(LlmapiAccuracyTestHarness):
     def test_nvfp4_multi_gpus(self, tp_size, pp_size, ep_size, mtp_nextn, fp8kv,
                               attention_dp, cuda_graph, overlap_scheduler,
                               max_batch_size, moe_backend):
-
-        kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.80)
+        kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.70)
         pytorch_config = dict(
             disable_overlap_scheduler=not overlap_scheduler,
             cuda_graph_config=CudaGraphConfig() if cuda_graph else None,
@@ -1830,7 +1808,7 @@ def test_fp8(self, tp_size, pp_size, ep_size, attention_dp, cuda_graph,
             task.evaluate(llm)
 
     @skip_pre_blackwell
-    @pytest.mark.skip_less_device(8)
+    @pytest.mark.skip_less_mpi_world_size(8)
     @pytest.mark.parametrize(
         "tp_size,pp_size,ep_size,attention_dp,cuda_graph,overlap_scheduler,moe_backend",
         [(8, 1, 8, True, True, True, "CUTLASS"),
@@ -1839,6 +1817,11 @@ def test_fp8(self, tp_size, pp_size, ep_size, attention_dp, cuda_graph,
     )
     def test_nvfp4(self, tp_size, pp_size, ep_size, attention_dp, cuda_graph,
                    overlap_scheduler, moe_backend):
+        if moe_backend == "TRTLLM":
+            pytest.skip(
+                "TRTLLM moe backend has accuracy issues: https://nvbugspro.nvidia.com/bug/5404726"
+            )
+
         pytorch_config = dict(
             disable_overlap_scheduler=not overlap_scheduler,
             cuda_graph_config=CudaGraphConfig() if cuda_graph else None,
diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt
index 35dcc590144..4b4c65cfcc7 100644
--- a/tests/integration/test_lists/waives.txt
+++ b/tests/integration/test_lists/waives.txt
@@ -416,7 +416,6 @@ test_e2e.py::test_ptp_quickstart_advanced_8gpus[Nemotron-Ultra-253B-nemotron-nas
 examples/test_multimodal.py::test_llm_multimodal_general[Qwen2-VL-7B-Instruct-pp:1-tp:1-float16-bs:1-cpp_e2e:False-nb:4] SKIP (https://nvbugs/5385981)
 examples/test_multimodal.py::test_llm_fp8_multimodal_general[fp8-fp8-cnn_dailymail-Qwen2-VL-7B-Instruct-pp:1-tp:1-bfloat16-bs:1-cpp_e2e:False] SKIP (https://nvbugs/5385987)
 examples/test_multimodal.py::test_llm_multimodal_general[Phi-4-multimodal-instruct-pp:1-tp:1-float16-bs:1-cpp_e2e:False-nb:1] SKIP (https://nvbugs/5385992)
-accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_tp8] SKIP (https://nvbugs/5377914)
 examples/test_multimodal.py::test_llm_multimodal_general[kosmos-2-pp:1-tp:1-float16-bs:1-cpp_e2e:False-nb:1] SKIP (https://nvbugs/5387422)
 examples/test_multimodal.py::test_llm_multimodal_general[fuyu-8b-pp:1-tp:1-float16-bs:1-cpp_e2e:False-nb:1] SKIP (https://nvbugs/5387424)
 test_e2e.py::test_ptp_quickstart SKIP (https://nvbugs/5387762)